Anthropic ended a commitment to not train or release risky models without commensurate safeguards, replacing it instead with a more narrow policy that only requires such restraint in particular cases, including if Anthropic is “in the lead” among competitors. In an announcement of the shift, ...